Short-Term Memory in Referring Expression Generation

PROJECT NUMBER: 43

AUTHOR: Will Culpepper, Electrical Engineering | MENTOR: Thomas Williams, Computer Science

ABSTRACT

—

Effective interfacing between humans and robots requires human-like natural language and dialogue capabilities. To engage in human-like dialogue, robots need to be able to describe objects in their environment; this capability is called “Referring Expression Generation.” As speakers repeatedly refer to similar objects, they tend to re-use the same properties used in previous descriptions, in part to help the listener, and in part to simply because those are the properties held in their short-term memory. The goal of this project is to investigate short-term memory models that will enable interactive robots to describe the world around them in an effective, natural manner. This project delves into different psychological models of “forgetting” and how it affects referring expression generation. In previous work, master’s student Kellyn Larson developed two models of short-term memory. In our work, we are designing a human-subject experiment to empirically evaluate those models. In the future, we’re hoping to apply these models to human-robot interaction to ensure a more naturally flowing conversation.

VISUAL PRESENTATION

—

AUTHOR BIOGRAPHY

—

Will Culpepper is a sophomore in electrical engineering seeking a minor in computer science. The short-term memory program is his first research project, and in the future he’d like to do research and develop projects in extraterrestrial robotics.

Sponsored by Phillips 66